Thief

Thief, a 2014 stealth title built on a modified Unreal Engine 3, presents an interesting benchmark scenario for the Intel Core Ultra 5 125H and NVIDIA GeForce RTX 5090 combination. The data shows a system that is heavily GPU-bound at higher resolutions, yet reveals a surprising CPU bottleneck at 1080p. The measured frame rates across the three resolutions and four settings presets provide a clear picture of how this modern mobile processor and flagship desktop graphics card interact with a last-generation game engine.

Resolution Scaling

The performance scaling from 1080p to 4K reveals the classic signature of a GPU-limited workload, but with a notable exception at the lower resolution. At 1080p with High settings, the system produces an average of 135 FPS. Moving to 1440p, that figure drops to 106 FPS, a 21.5% reduction. Scaling further to 4K, the average falls to 62 FPS, representing a 54.1% decrease from the 1440p result and a 54.1% decrease from the 1080p result. This steep decline with increasing pixel count indicates that the RTX 5090 is being tasked progressively harder, which is expected given the 4K resolution demands 4 times the pixels of 1080p.

However, the scaling pattern is not uniform across all settings. The Low preset shows a much flatter curve. At 1080p Low, the system achieves 164 FPS. At 1440p Low, it only drops to 145 FPS, an 11.6% reduction. At 4K Low, the average is 108 FPS, a 34.1% reduction from 1440p but still a very playable number. This smaller performance delta suggests that at Low settings, the game’s visual complexity is reduced enough that the CPU can begin to play a more significant role, particularly at 1080p where the GPU has the least amount of work to do. The 1080p Low result being only 11.6% faster than the 1440p Low result, compared to the 21.5% gap at High settings, hints that the CPU’s 14 cores and 18 threads are reaching their limit in feeding the GPU frames at lower graphical loads.

The Medium preset offers the most complete data set, including minimum and maximum frame rates. At 1080p Medium, the average is 149 FPS with a minimum of 126 FPS and a maximum of 171 FPS. At 1440p Medium, the average drops to 131 FPS, with a minimum of 112 FPS and a maximum of 151 FPS. At 4K Medium, the average falls to 82 FPS, with a minimum of 70 FPS and a maximum of 94 FPS. The variance between minimum and maximum FPS is relatively consistent, suggesting stable frame pacing without severe hitches. The scaling from 1080p to 1440p is an 12.1% drop, while moving from 1440p to 4K is a 37.4% drop, reinforcing that the GPU becomes the primary limiting factor as resolution increases.

The data indicates that at 1080p, the system is not purely GPU-bound, as the performance differences between Low and High settings are compressed compared to higher resolutions. For example, at 1080p, the difference between Low and High is 17.7%, but at 4K, the difference is 42.6%. This suggests that the CPU’s performance ceiling is being approached at 1080p, preventing the GPU from fully stretching its legs. The RTX 5090’s massive compute resources are somewhat held back by the Ultra 5 125H at the lowest resolution, but at 1440p and 4K, the GPU’s rendering workload takes over as the decisive factor.

GPU Role

The NVIDIA GeForce RTX 5090 is the dominant component in this pairing, and its specifications align with the observed performance data. With 32 GB of GDDR7 memory on a 512-bit bus, the GPU has a memory bandwidth of 1.79 TB/s. This is a massive amount of bandwidth, and for a game like Thief, which was released in 2014, it is far more than sufficient. The game’s textures and assets are unlikely to stress the 32 GB frame buffer, even at 4K Ultra settings. The 108 FPS average at 4K Low and 62 FPS at 4K High show that the GPU’s raw shader throughput, measured at 104.8 TFLOPS FP32, is the deciding factor for these framerates, not memory capacity.

The GPU’s clock speeds, with a base of 2017 MHz and a boost of 2407 MHz, are high for a desktop-class card. The performance difference between the Low and Ultra presets at 4K is stark: 108 FPS versus 59 FPS. This 45.4% performance gap is directly attributable to the increased shading and texture work demanded by higher settings. The RTX 5090’s 21,760 shading units and 680 texture mapping units are being utilized more heavily as the preset quality increases, but the game engine’s age means it cannot fully leverage the card’s newer architectures, such as Blackwell 2.0’s dedicated RT and Tensor cores.

The 4K Ultra result of 59 FPS is a notable data point. It shows that even the most powerful consumer GPU on the market, with a 92.2 billion transistor count and a 750 mm² die, cannot maintain a 60 FPS average in this game at maximum settings. This is not a limitation of the GPU’s hardware but rather a reflection of the game’s engine and its draw call efficiency. The benchmark results indicate that the GPU’s raw power is being constrained by the game’s rendering path, which was designed for hardware from a decade prior. The RTX 5090 is effectively running at a fraction of its potential, as evidenced by its 92nd percentile ranking among all GPUs, yet it still cannot brute-force its way past the engine’s limitations at 4K Ultra.

The GPU’s role is clear: it provides the necessary headroom to run the game at very high frame rates at 1080p and 1440p, but its performance at 4K is capped by the game’s single-threaded draw call bottlenecks and lack of modern API features. The 3DMark Steel Nomad DX12 score of 18,355 is irrelevant to Thief’s performance, as the game uses an older DirectX 11 API, which is reflected in the PassMark DirectX 11 score of 341. This older API limits the GPU’s ability to parallelize work efficiently, making the game more sensitive to raw clock speed and memory bandwidth than to the card’s advanced features.

Settings Recommendations

Based on the measured FPS data, the optimal settings for this combination depend heavily on the target resolution and desired frame rate. At 1080p, the Ultra preset delivers an average of 132 FPS, which is only 2.2% lower than the High preset’s 135 FPS. Given the minimal performance difference, the Ultra preset is the clear recommendation for 1080p, as it provides the highest visual fidelity without a meaningful cost to frame rate. The Low preset at 1080p offers 164 FPS, but this is unnecessary for most displays, and the visual trade-off is substantial.

At 1440p, the High preset achieves 106 FPS, while the Ultra preset achieves 101 FPS. The 4.7% difference between them is small, and the Ultra preset is again the better choice for a balanced experience. The Medium preset at 131 FPS is an option for players with high-refresh-rate 1440p monitors, but the jump to High is only a 19.1% reduction in frame rate, making High a better visual compromise. The Low preset at 145 FPS is not recommended, as the visual quality degradation is not worth the 36.8% improvement over High.

At 4K, the situation changes. The Ultra preset produces 59 FPS, which is just below the 60 FPS target. The High preset at 62 FPS is only 4.8% faster, so it does not provide a substantial improvement. The Medium preset, however, delivers 82 FPS, which is a 32.3% improvement over Ultra. This is the most significant jump in performance between consecutive presets. For a smooth 4K experience, the Medium preset is the recommended choice, as it provides a comfortable margin above 60 FPS while retaining a reasonable level of visual detail. The Low preset at 108 FPS is an option for those prioritizing frame rate above all else, but the Medium preset offers a better balance for the 4K resolution.

The data suggests that the game scales well with settings at 4K, allowing for fine-tuning. The Medium preset’s minimum FPS of 70 at 4K ensures that the frame rate rarely dips below the 60 FPS threshold, making it the most consistent option. In summary, the best experience per the measured rows is Ultra at 1080p, Ultra at 1440p, and Medium at 4K.

FAQ

Q: What is the average frame rate at 1080p with Ultra settings?

A: The Intel Core Ultra 5 125H and NVIDIA GeForce RTX 5090 combination achieves an average of 132 FPS at 1920x1080 with the Ultra preset.

Q: Is the system capable of 4K gaming at 60 FPS?

A: Yes, but only at Medium settings. The 4K Medium preset yields an average of 82 FPS, while the High and Ultra presets produce 62 FPS and 59 FPS respectively, with Ultra falling just short of the 60 FPS mark.

Q: How does the CPU’s performance compare to its nearest rivals?

A: The Intel Core Ultra 5 125H has an average benchmark score of 27,507, which is 0.1% lower than the AMD Ryzen 7 5800 and 0.3% lower than both the Intel Core i5-12600K and AMD Ryzen 7 5700X. It is 0.5% faster than the AMD Ryzen 3 PRO 5355G.

Q: What is the minimum FPS recorded for this system?

A: The only measured minimum FPS values are for the Medium preset. The lowest minimum is 70 FPS at 3840x2160, followed by 112 FPS at 2560x1440, and 126 FPS at 1920x1080.

Q: Does the GPU have enough memory for Thief?

A: The GPU has 32 GB of GDDR7 memory. Thief, being a 2014 game, is not expected to use a significant portion of this capacity, so memory size is not a limiting factor in the measured results.

Q: How does the GPU’s overall benchmark score rank?

A: The NVIDIA GeForce RTX 5090 has a PassMark G3D score of 39,650, placing it in the 92nd percentile of all GPUs. Its average benchmark score is 79,842, which is 0.3% higher than the NVIDIA Tesla P100 PCIe 16 GB and 0.6% higher than the NVIDIA Tesla P100 PCIe 12 GB.

How This Combo Ranks

The combination of the Intel Core Ultra 5 125H and NVIDIA GeForce RTX 5090 ranks 694th out of 2,851 tested combos for Thief. This places the system in the top 24.3% of all configurations tested. This ranking is surprisingly low for a system with a top-tier GPU, which suggests that the pairing is suboptimal for this specific game. The bottleneck analysis from the resolution scaling confirms this, as the CPU’s performance ceiling at 1080p prevents the system from achieving the extremely high frame rates that the GPU alone could deliver.

The rank of 694 is a consequence of the CPU’s position in the benchmark hierarchy. The Ultra 5 125H has a percentile rank of 79 among all CPUs, indicating it is a capable processor, but it is not in the top tier for gaming. Its nearest rivals, such as the AMD Ryzen 7 5800 and Intel Core i5-12600K, have similar average scores, but these are desktop processors with different thermal and power characteristics. The 125H’s mobile design, with a 28 W TDP, is likely the reason for its performance ceiling in gaming workloads. The game’s engine, which is not optimized for many-core processors, may not be able to fully utilize the 14 cores and 18 threads, leaving the system at a disadvantage compared to combos with higher single-thread performance.

Despite the low rank, the system still delivers playable frame rates at all resolutions. The 4K Medium preset’s 82 FPS average is a solid result, and the 1080p Ultra result of 132 FPS is excellent. The rank reflects the system’s inability to push into the extreme high frame rate territory that other high-end combos might achieve, but it does not indicate a failure to provide a good gaming experience. The data shows that the RTX 5090 is the primary driver of performance, but its potential is somewhat muted by the CPU, resulting in a combo that is strong but not class-leading for Thief.

CPU Role

The Intel Core Ultra 5 125H, with its 14 cores (6 performance and 8 efficient) and 18 threads, plays a secondary but critical role in this system’s performance. Its base clock of 3.60 GHz and boost clock of 4.50 GHz are moderate for a modern processor. The CPU’s Cinebench R23 multi-core score of 12,804.5 and single-core score of 1,647.5 provide context for its capabilities. The single-core score is particularly relevant for gaming, as many games, including Thief, rely heavily on single-threaded performance for game logic and draw calls.

The benchmark data reveals that the CPU is a limiting factor at 1080p. The fact that the 1080p Low preset (164 FPS) is not significantly faster than the 1080p Medium preset (149 FPS) suggests that the CPU is unable to feed the GPU with enough frames to fully utilize its potential. The difference between these two presets is only 10.1%, which is a small gap compared to the 4K Low to Medium difference of 24.1%. This indicates that the CPU’s processing power is the bottleneck at lower resolutions, preventing the GPU from rendering more frames.

The CPU’s cache structure, with 112 KB of L1 cache per core, 2 MB of L2 cache per core, and 18 MB of shared L3 cache, is substantial. However, the game’s engine may not be optimized to take advantage of this, and the mobile platform’s memory bandwidth of 89.6 GB/s could be a limiting factor. The CPU’s PassMark single-thread score of 3,399 is decent, but it is not exceptional. In comparison to its rivals, the CPU’s average benchmark score is statistically tied with several desktop processors, but the mobile form factor’s thermal and power constraints likely prevent it from sustaining high clocks for extended periods.

The CPU’s role in the 4K results is less prominent. At 4K, the GPU is clearly the bottleneck, as the performance scales directly with the GPU’s workload. The CPU has enough headroom to keep up with the GPU’s rendering demands, as evidenced by the 82 FPS at 4K Medium. The CPU’s 18 threads are likely sufficient to handle the game’s background tasks, but the game’s primary thread is still a constraint. The data indicates that for this game, the CPU’s performance is adequate for 1440p and 4K gaming, but it holds back the system at 1080p, preventing the RTX 5090 from achieving its full potential.

Measured FPS Breakdown

The following table details the exact average frame rates measured for each resolution and settings combination.

| Resolution | Settings | Avg FPS | Min FPS | Max FPS |

| :--------- | :------- | :------ | :------ | :------ |

| 1920x1080 | Low | 164 | - | - |

| 1920x1080 | Medium | 149 | 126 | 171 |

| 1920x1080 | High | 135 | - | - |

| 1920x1080 | Ultra | 132 | - | - |

| 2560x1440 | Low | 145 | - | - |

| 2560x1440 | Medium | 131 | 112 | 151 |

| 2560x1440 | High | 106 | - | - |

| 2560x1440 | Ultra | 101 | - | - |

| 3840x2160 | Low | 108 | - | - |

| 3840x2160 | Medium | 82 | 70 | 94 |

| 3840x2160 | High | 62 | - | - |

| 3840x2160 | Ultra | 59 | - | - |

The data shows a clear hierarchy. At 1080p, the system delivers playable frame rates across all presets, with the Low preset providing a 24.2% improvement over Ultra. At 1440p, the performance is still strong, with the Low preset achieving a 43.6% improvement over Ultra. At 4K, the gap between Low and Ultra widens to 83.1%, demonstrating the GPU’s workload scaling. The Medium preset is the only one with measured minimum and maximum FPS, and it consistently provides a stable experience, with the minimum FPS at 4K being 70, which is above the 60 FPS threshold. The Ultra preset at 4K is the only measured configuration that falls below 60 FPS, with an average of 59 FPS.